As I said, when you first start on this, you kind of go with what's easiest, because you're learning. You're bringing your organization along as much as you're bringing the public along.
It's really what's available. As I think many of us said today, don't start with the hardest, most politically sensitive data. Start with what's easy. But then, quite quickly, I think we all need to move to an assumption that all data will become public unless there are compelling reasons not to release it. Then you start to build your data sets in a whole different way.
Really, the criteria include, first and foremost, whether it is sensitive at a personal level. Will it confound privacy legislation and respect it? Is it interesting, and does the public want it?
Those are probably the most important criteria once you're past that.